How to Read SEC Filings (10-K, 10-Q, 8-K) With AI
SEC filings are the most reliable public information about a US-listed company. They are also long, dense, and easy to skip. AI can make them readable, but only if you use it to read the filing, not to replace it. This guide covers which filings matter, where to find them, and how to use AI on them safely.
The filings that matter to investors
| Filing | What it is | When it appears |
|---|---|---|
| 10-K | Annual report: business, risks, audited financial statements | Once a year, after the fiscal year ends |
| 10-Q | Quarterly report: unaudited financials and updates | After each of the first three fiscal quarters |
| 8-K | Current report on a material event (results, deals, executive changes) | Within days of the event |
| DEF 14A | Proxy statement: executive pay, board, shareholder votes | Before the annual meeting |
| Form 4 | Insider buying and selling by officers and directors | Shortly after the trade |
| 13F | Quarterly holdings of large investment managers | After each quarter |
For a long-term holder, the 10-K and 10-Q do most of the work. 8-Ks tell you when something changed between reports.
Where to find them
Every filing is free on the SEC’s EDGAR system at sec.gov. Search by company name or ticker, then filter by form type. EDGAR also offers full-text search across filings, which is useful for finding every mention of a customer, product or risk.
What to read first in a 10-K
You do not need to read all of it. Start here:
- Item 1, Business. How the company actually makes money, by segment.
- Item 1A, Risk Factors. Long and partly boilerplate, but new or reworded risks are a signal.
- Item 7, Management’s Discussion and Analysis (MD&A). Management explains the numbers. Compare it with last year’s.
- Item 8, Financial Statements and notes. Income statement, balance sheet, cash flow. The notes hold debt terms, segment data and accounting choices.
The 10-Q follows a shorter version of the same structure for each quarter.
How to use AI on filings without being misled
Large language models are good at summarising long documents and bad at admitting when they do not know. Four rules keep them honest.
Give it the document
Do not ask a chatbot what a company’s 10-K says. Download or paste the relevant section and ask about that text. Otherwise it answers from training data that may be out of date or wrong.
Require verbatim quotes
Ask for the exact sentence behind every claim, then spot-check a few against the filing. If the quote is not there, discard the claim. Fabricated quotes are a known failure mode of language models.
Ask for change, not summary
The most valuable question is usually what changed. Compare this year’s risk factors or MD&A with last year’s and ask what was added, removed or softened.
Ask for the opposing case
A summary tends to repeat management’s framing. Ask explicitly for the bear case using only the filing, and for the numbers that contradict the narrative, such as operating cash flow lagging net income.
Prompts to start with
- “Using only the Item 1A text below, list risks that are new compared with last year’s list. Quote each one.”
- “From these financial statements, show revenue, operating income and operating cash flow for each year in a table. Flag any year where cash flow and income move in opposite directions.”
- “Summarise this 8-K in three sentences and say what it changes for an existing shareholder.”
Automating it
Reading filings for every holding, every quarter, is where most investors give up. Quorum, built by Natkal, automates this step for a personal portfolio. A recursive reading model works through 10-K and 10-Q filings from SEC EDGAR turn by turn and extracts verbatim quotes that are verified against fabrication. Those quotes then feed an eleven-seat AI council that debates each holding. Every verdict shows the evidence it rests on and the dissent it overcame.
For how that debate works, see AI Hedge Fund Agents: How Multi-Agent Investing Works. To compare AI research tools, see Which AI Is Best for Stock Analysis?.